Low Power MAC Protocols for Infrastructure Wireless Sensor Networks
A. El-Hoiydi, J.-D. Decotignie and J. Hernandez
CSEM, Swiss Center for Electronics and Microtechnology, Inc.
Rue Jaquet-Droz 1, 2007 Neuchâtel, Switzerland
e-mail: {aeh, jdd, jhz}@csem.ch
Abstract: This paper addresses low power MAC pro-
tocols for the downlink of infrastructure wireless sensor
networks. We are interested in the trade-off between power
consumption and transmission delay, focusing on low traf-
fic. We describe WiseMAC (Wireless Sensor MAC), a new
protocol for the downlink of infrastructure wireless sen-
sor networks. Another original contribution is the anal-
ysis of the performance of PTIP (Periodic Terminal Initi-
ated Polling). Here, polling is used in the reversed direc-
tion as compared to common polling protocols. WiseMAC
and PTIP are compared with PSM, the power save proto-
col used in both the IEEE 802.11 and IEEE 802.15.4 Zig-
Bee standards. Analytical expressions are given for the
power consumption and the transmission delay for each
protocol, as a function of the wake-up period. It is shown
that WiseMAC provides, for the same delay, a significantly
lower power consumption than PSM. Although less energy
efficient than WiseMAC and PSM, it is shown that PTIP
can, thanks to its implementation simplicity, become at-
tractive for applications tolerating large transmission de-
lays1 .
1 Introduction
Inexpensive integrated system-on-a-chip devices
comprising a radio transceiver and a microcontroller
have been since a few years a subject of research [1, 2].
Industry is now selling such devices [3]. They will
permit to implement ubiquitous computing applications
where small battery powered nodes are interconnected
via a wireless network. As now widely recognized, one
of the main issues is the power consumption of such
devices.
1.1 Problem Statement
This paper discusses the power consumption of
medium access control protocols in an infrastructure
wireless sensor network, focusing on low downlink traf-
fic. The term ”sensor” is used to emphasize the low
power requirement. The mobile nodes considered in this
paper may be sensors (e.g. fire alarm), but also other
kinds of devices such as actuators, personal digital assis-
tants, etc.
Unlike most research dealing with wireless sensor net-
works, we do not consider an ad hoc multi-hop network
topology, but an infrastructure network. An infrastruc-
ture network is composed of a number of access points
interconnected through a backbone network. Each ac-
cess point is serving a number of sensor nodes. Such a
topology can be envisaged for example in smart build-
ing applications, where the Ethernet or powerline ca-
bling can be used for the backbone network. The main
1 The work presented in this paper was supported in part by the
National Competence Center in Research on Mobile Information and
Communication Systems (NCCR-MICS), a center supported by the
Swiss National Science Foundation under grant number 5005-67322.
characteristic of access points is that they are assumed
to be energy unconstrained. In this sense, this work can
also apply to clustered ad hoc networks with solar pow-
ered cluster heads, as proposed in [4]. Finally, one can
imagine a vehicle mounted mobile access point moving
through an cloud of sensors to collect data.
An energy efficient wireless MAC protocol should
minimize the four sources of energy waste [5]: idle lis-
tening, overhearing, collisions and protocol overhead.
Idle listening refers to the active listening to an idle chan-
nel, waiting for a potential packet to arrive. Overhearing
refers to the reception of a packet, or of part of a packet,
that is destined to another node. Collisions should of
course be avoided as retransmissions cost energy. Fi-
nally, protocol overhead refers to the frame headers and
the signalling required by the MAC protocol.
As the power consumption of a transceiver in receive
mode is far from being negligible, idle listening can be-
come the main source of energy waste, especially in low
traffic conditions. To reach a low average power con-
sumption, the transceiver must be shut down part of the
time (i.e. duty cycling).
In infrastructure networks, one must distinguish the
downlink (access point to sensor nodes) from the uplink
(sensor nodes to access point). In the downlink direction,
the challenge is to transmit data from the access point to
some sensor node, without requiring the sensor node to
continuously listen to the channel. The MAC protocol
must mainly mitigate idle listening and overhearing on
the sensor nodes. The problem is different in the uplink
direction. As the access point is not energy limited, it
can listen all the time to the channel. The uplink MAC
protocol requires no wake-up scheme. The issue to re-
solve in the uplink direction is the multiple access to a
shared medium. If the system is operated near capacity,
this problem is very complex. However, if only a mod-
erate traffic is present on the channel, the simple non-
persistent CSMA protocol [6] can clearly approach the
ideal case, with no idle listening, no overhearing and lit-
tle collisions. In this paper, we will therefore focus on
the downlink problem.
Sensor networks are usually meant for the acquisi-
tion of data, either periodically or based on events (e.g.
alarms). The uplink traffic can be expected to be high,
at least during certain periods. On the other hand, the
downlink is foreseen to carry configuration and query-
ing traffic. With such a traffic, inter-arrivals measured
in minutes or hours will be common. We will assume
throughout this paper that the inter-arrival between pack-
ets is much larger than the time needed to transmit a
packet.
1.2 Original Contributions
This paper contains two original contributions. The
first one is the proposal of WiseMAC (Wireless Sensor
MAC) for the downlink of infrastructure wireless sen-
sor networks. This protocol has been proposed in [7]
for multi-hop wireless sensor networks. We show here
that it is also of interest for infrastructure wireless sen-
sor networks. The second contribution is the analysis of
the Periodic Terminal Initiated Polling protocol (PTIP).
Polling protocols are usually used to poll mobile stations
from a central access point in order to avoid collisions in
the uplink direction [8]. In this paper, we analyze the re-
versed usage of polling, for the downlink direction. Such
a usage of polling has received little attention from the
research community, because it is very inefficient in high
traffic conditions. We show that this simple protocol can
become attractive in certain conditions. WiseMAC and
PTIP are compared with the power save protocol used in
IEEE 802.11 [9] and IEEE 802.15.4 ZigBee [10] and to
an hypothetic ideal protocol.
1.3 Related Work
A large effort has been devoted by the research com-
munity to the development of medium access control
protocols for wireless computer networks [6, 11]. Such
protocols have been primarily designed to minimize the
delay and to maximize the throughput. The power con-
sumption has only later become an issue of large inter-
est. In [12], a comparison between the power consump-
tion of several wireless MAC protocols is given. In their
analysis, the authors focused on high traffic conditions.
In this paper, we focus, in the contrary, on low traffic
conditions. Low traffic is expected to be very common
in many applications of ubiquitous computing, where a
very long lifetime is required. It is therefore necessary
to understand and minimize the energy consumption of
MAC protocols in low traffic conditions.
Research on ultra low power MAC protocols for ad
hoc wireless sensor network has started a few years ago.
A number of proposals are available, from which one
can cite S-MAC [5], Piconet [13] and wake-up radio [2].
Protocols for ad hoc sensor networks can potentially be
interesting for the downlink of infrastructure sensor net-
works. For example, the Piconet protocol can be seen as
a distributed version of PTIP. The wake-up radio scheme
would certainly be of interest for the downlink of infras-
tructure networks, if a wake-up radio receiver hardware
consuming almost nothing becomes a reality. The work
presented in this paper differs from previous research on
MAC protocols for sensor networks, mainly because we
focus on an infrastructure topology, and investigate how
the unconstrained energy supply of the access point can
be exploited.
Another field related to infrastructure sensor networks
is the field of paging systems. Several standards have
been developed over the years, from which POCSAG
and FLEX [14, 15]. The techniques used by paging sys-
tems are useful inspiration sources, but these protocols
cannot be used as such for infrastructure wireless sen-
sor networks. Paging protocols seek the capacity-energy
optimum, while we seek the delay-energy optimum.
1.4 Paper Organization
The rest of the paper is organized as follows: Section 2
describes the mentioned low power MAC protocols. The
power consumption and the delay of these protocols are
given in Section 3. A performance comparison is made
in Section 4. Section 5 contains the main findings from a
sensitivity analysis. Section 6 gives concluding remarks.
2 Low Power Downlink MAC Protocols
Before proceeding to the description of the protocols,
let us define models for the radio transceiver and the traf-
fic.
2.1 Radio Model
When considering low power protocols, it is very im-
portant to model precisely the transition delays between
the different states of a transceiver and the power con-
sumption in those states. The following states can be
identified:
DOZE The transceiver is not able to transmit nor re-
ceive, but is ready to quickly power on into the re-
ceive or the transmit state,
RX The transceiver is listening to the channel (receiving
data or trying to demodulate data out of a noisy or
idle channel),
TX The transceiver is transmitting data.
The power consumed in these states will be denoted
as PZ , PR and PT . To simplify analytical expressions,
we define PbR = PR −PZ as the increment in power con-
sumption caused by being in the RX state (as compared
to the DOZE state), and PbT = PT − PZ as the increment
in power consumption caused by being in the TX state.
We denote with TS the setup time required to turn on
the transceiver into the RX state, starting from the DOZE
state. We finally denote with TT the turn-around time
which is required to switch the transceiver between the
RX and TX states. During the state change phases, we
assume that the transceiver consumes a power PR , as
all electronics is powered on at the exception of the last
stage RF power amplifier.
2.2 Traffic Model
We consider a population of N sensor nodes under
the responsibility of one access point. Downlink Poisson
traffic arrives at the access point from the fixed network
λ
at a global rate λ. We assume that an equal part N of this
traffic is destined to each sensor node. A given sensor
node will receive packets with an average packet inter-
arrival time of L = N λ.
Data packets have a constant duration TD . Control
packets (pollings, acknowledgements, traffic indication
map beacons) have a constant duration TC .
As mentioned in the introduction, we assume a low
traffic, where the inter-arrival 1/λ is much larger than
the sum of the durations of a data packet, of the turn-
around and of a control packet:
Low Traffic Assumption: 1/λ À TD + TT + TC
 Arrival, wait for transmit
right moment
ACCESS WAIT W DATA
POINT
TP
SENSOR TW
NODE
Wake up, Wake up, Wake up,
T TT TC
medium idle medium idle medium busy, D
DOZE RX TX receive message
Figure 1: WiseMAC - Principle of Operation
2.3 WiseMAC
WiseMAC is based on the preamble sampling tech-
nique. This technique consists in periodically sampling
the medium to check for activity. By sampling the
medium, we mean listening to the radio channel for a
short duration, e.g. the duration of a modulation sym-
bol. All sensor nodes in a network sample the medium
with the same constant period TW . Their relative sam-
pling schedule offsets are independent. If the medium
is found busy, a sensor node continues to listen until a
data packet is received or until the medium becomes idle
again. At the access point, a wake-up preamble of dura-
tion equal to the medium sampling period is transmitted
in front of every data frame to ensure that the receiver
will be awake when the data portion of the packet will
arrive. This technique provides a very low power con-
sumption when the channel is idle. See [16] and [17] for
details. The disadvantages of this protocol are that the
(long) wake-up preambles cause a throughput limitation
and a large power consumption overhead in reception.
The overhead in reception is not only born by the in-
tended destination, but also by all other nodes overhear-
ing the transmission.
The novel idea introduced by WiseMAC consists in
letting the access point learn the sampling schedule of
all sensor nodes. Knowing the sampling schedule of the
destination, the access point starts the transmission just
at the right time with a wake-up preamble of minimized
duration TP as illustrated in Fig. 1. The access point
keeps an up-to-date table with the sampling schedule of
all sensor nodes. The sampling schedule information is
gained through the inclusion in every acknowledgement
packet of the remaining time until the next scheduled
sampling.
The duration of the wake-up preamble must be com-
puted such as to compensate for the drift between the
clock at the access point and on the sensor nodes.
This drift is proportional to the time since the last re-
synchronization (i.e. the last time an acknowledgement
was received from a given sensor node). Let θ be the
frequency tolerance of the time-base quartz, TW be the
interval between preamble samplings and L the interval
between two communications. The required duration of
the wake-up preamble is
TP = min(4θL, TW )
This expression is obtained as follow: Assume that
the access point has received fresh timing information
from some sensor node at time 0, and that the access
point wants to send a packet to this sensor node at the
sampling time L. If the sensor node quartz has a real
frequency of f (1 + θ) instead of f , its clock will have
an advance of θL at time L. It is hence needed to start
ACK the preamble θL in advance. Because the clock of the
access point might be late, it must target a time 2θL in
advance. Because the clock of the access point might be
early, and the clock of the sensor node late, the duration
of the wake-up preamble must be of 4θL. If L is very
large, 4θL may be larger that the sampling interval TW .
In those cases, a preamble of duration TW is used.
If the traffic is high, the interval L between transmis-
sions will be small, and so the wake-up preamble (4θL).
If the traffic is low, the interval between transmissions
will be high, but at maximum equal to TW . This impor-
tant property makes the WiseMAC protocol adaptive to
the traffic. The per-packet overhead decreases in high
traffic conditions.
Overhearing is naturally mitigated when the traffic is
high, thanks to the combined use of the preamble sam-
pling technique and the minimization of the wake-up
preamble duration. As already mentioned, sensor nodes
are not synchronized among themselves. Their relative
sampling schedule offsets are independent. In high traf-
fic conditions, the duration of the wake-up preamble be-
ing smaller than the sampling period, short transmission
are likely to fall in between sampling instants of poten-
tial overhearers.
Finally, it is interesting to note that collisions are not
possible using WiseMAC for a downlink channel, as the
access point is the only initiator of transmissions.
2.4 Periodic Terminal Initiated Polling - PTIP
With the PTIP protocol, the access point buffers
downlink traffic. Sensor nodes regularly send a POLL
packet to the access point to get potentially buffered data.
The access point replies with a DATA packet if one was
buffered, or with a (shorter) control packet if the queue
for the requesting node was empty. To mitigate colli-
sions between sensor nodes, POLL packets are sent us-
ing the CSMA protocol. To avoid systematic contentions
between synchronized nodes, the time interval between
POLL transmissions is a random variable with mean
value TW . The basic principle of operation of PTIP is
illustrated in Fig. 2.
If the response to the POLL is correctly received, the
sensor node goes back to sleep until the next scheduled
polling time. With PTIP, it is not required to send an ac-
knowledgement when a downlink DATA packet has been
received correctly. Instead, the POLL is repeated until a
response is received correctly. To let the access point
know whether a DATA packet has to be retransmitted or
not, the sequence number of the last correctly received
DATA packet is piggy-backed on every POLL packet.
2.5 IEEE 802.11/802.15.4 Power Save Mode - PSM
A power save mode (PSM) has been specified in the
IEEE 802.11 standard to permit a lower power consump-
(1)
tion at the cost of a larger delay [9]. The same scheme
has been selected for the newer IEEE 802.15.4 ZigBee
standard [10]. The access point buffers incoming traffic.
ACCESS NO Arrival sensor node that it must poll the access point to down-
DATA DATA
POINT load the following packet. This scheme permits to use
a wake-up interval that is larger than the average inter-
SENSOR
TW
val between the arrivals for a given node (TW > L). It
POLL POLL
NODE permits to reduce the queuing delay at the access point,
TS TC TT TD especially in the event of traffic bursts.
DOZE RX TX
3 Power Consumption and Delay
Figure 2: Periodic Terminal Initiated Polling (PTIP) - Principle
of Operation This section introduces analytical expressions to com-
pute the power consumption and the delay of WiseMAC,
Arrival PTIP and PSM, under the low traffic assumption (1/λ À
ACCESS
POINT
TIM TIM DATA TD + TT + TC ). Due to space limitations, little expla-
nation is given on the derivation of those expressions.
Interested readers are referred to [18] for details.
SENSOR TW
NODE
POLL It can be shown that the average power consumed by
WiseMAC, PTIP and PSM is respectively given by
DOZE RX TX
TS TCDC TC TT TC TT TD

Figure 3: Optimized Power Save Mode (PSM) - Principle of PbR (TS + 1/B)
PW iseM AC = PZ +
Operation TW
PbR (TP /2 + TD + TT ) + PbT TC
+
L
2
A beacon is periodically transmitted with period TW . (T P + TD )
+ PbR (N − 1) (2)
This beacon contains the traffic indication map (TIM), 2LTW
which lists the sensor nodes for which data packets have
been buffered. All sensor nodes wake-up regularly to re- TW PbT TC + PbR (TS + TT + TC )
ceive the TIM. If they discover their address in the TIM, PP T IP = PZ + e− L
TW
they poll the access point to receive the buffered data.
PbT TC + PbR (TS + TT + TD )
The standard requires the access point to reply to a + (3)
L
POLL after a given delay (10 µs in DS 802.11b). In
practice, it is difficult for the access point software to find
the right packet and prepare it for transmission within PbR (TS + TC )
PP SM = PZ + 2θPbR +
the specified delay. Instead, the access point replies to TW
the POLL with an ACK. This instructs the sensor node PbT TC + PbR (TD + 2TT )
+ (4)
to remain in listening mode. As soon as possible, the L
access point sends the DATA packet, which in then ac- Expression (2) is composed of the power consumed in
knowledged back by the sensor node. In summary, the the DOZE state, of the power consumption increments
polling procedure is composed of four packet transmis- caused by the preamble sampling activity (wake-up and
sions: POLL-ACK-DATA-ACK. In this paper, we are in- sense the channel during one radio symbol every TW ),
terested in the basic performance of protocols that would the reception of the packet (listen in average to half
use a traffic indication map. For a fair comparison with of the wake-up preamble, receive the data, turn-around
the other protocols, we consider a version of the PSM the transceiver and send the acknowledgement) and the
protocol, that is fully optimized for low power operation. overhearing of this packet by the N − 1 other sensor
We assume first that access point replies to a POLL with nodes ((TP + TD )2 /2TW is the average duration during
a DATA packet (after the needed delay), and secondly which a transmission of duration TP + TD is overheard
that a DATA packet is not acknowledged (as in PTIP, by some node sampling the medium with period TW , see
the acknowledgement is piggy-backed on the following [18]).
POLL). This procedure is illustrated in Fig. 3. Note that In expression (3), the first term represents the power
sensor nodes must listen to the channel TCDC = 2θTW consumed in DOZE state, the second term the cost of
TW
before the expected start of the TIM packet to compen- useless pollings (e− L is the probability of having no
sate for a potential drift between the access point clock arrivals in between two pollings) and the third the cost
and their clock. of downloading buffered packets.
2.6 More bit In expression (4), the first term represents the power
A very important detail shared by WiseMAC, PTIP consumed in DOZE state. The second term, 2θPbR , ac-
and PSM is the presence of a more bit in the header of counts for the time spent listening to the channel to cover
data packets. When this bit is set to 1, this indicates that the drift between the access point clock and the sen-
more data packets destined to the same sensor node are sor node clock (see [18]). The third term represents the
waiting in the access point buffer. With WiseMAC, this power consumed to power-on and listen to the beacon of
indicates to the sensor node that it must continue to listen duration TC every TW seconds. Finally, the fourth term
after having sent the acknowledgement. The next packet accounts for the transmission of poll packets and the re-
will follow. With PTIP and PSM, this indicates to the ception of data packets, including turning the transceiver
around before the POLL transmission and the DATA re- Param. Value Param. Value
ception. PZ 5 µW TD (50 bytes) 16 ms
The transmission delay is defined as the time elapsed PR 1.8 mW TC (10 bytes) 3.2 ms
between the arrival of a packet at the access point and the PT 27 mW N 10
end of its transmission to the destination sensor node. TS 0.8 ms L 1000 s
TT 0.4 ms θ 30 · 10−6
Under the low traffic assumption, it can be shown that
B 25 kbps
the transmission delay with WiseMAC, PTIP and PSM
is respectively given by Table 1: System Parameters

DW iseM AC = TW /2 + TP + TD (5)
60
Ideal
WiseMAC
DP T IP = TW /2 + TT + TD

Power [µW]
(6) 40 PTIP
PSM

20
DP SM = TW /2 + 2TC + 2TT + TD (7)
Expression (5) is composed of the average time be- 0 −2 −1 0 1 2 3
10 10 10 10 10 10
tween the arrival of a packet at the access point and the 3
TW [s]
10
start of the preamble transmission, the time to transmit 2
10
the preamble and the data.

Delay D [s]
1
10
Expression (6) is composed of the average time be- 10
0

tween the arrival of a packet and the end of the POLL 10

−1

reception, the time to turn-around into transmit state and 10

−2

−2 −1 0 1 2 3
to transmit the data packet. 10 10 10
TW [s]
10 10 10

Expression (7) is composed of the average time be-

tween the arrival of a packet and the start of the TIM
transmission, the time to transmit the TIM and the time
for the sensor node to send a POLL and receive the data
packet.
In the case of PTIP, we assume no collisions between
POLL packets. This implies that the sum of the data and
polling traffic must be small. In addition to the low traffic
assumption, we need to set TW À N TC for expressions
(3) and (6) to be valid.
In the case of PSM, when the TIM broadcast period
TW is chosen to be larger than the global inter-arrival
1/λ, several sensor nodes will enter in contention to
download the buffered packets. The computation of the
duration of the collision resolution interval with non-
persistent CSMA is unfortunately a problem for which
no analytical solution is available. Solutions exist in the
steady state [6] or when using more complex protocols
such as the binary tree collision resolution protocol [19].
We do not consider such results in this paper for space
limitation reasons and because we do not need to con-
sider collisions to compare these protocols in a useful
manner. To avoid the problem of potential collisions, we
consider with PSM a traffic sufficiently low such that at
most one packet is received in average in a period TW .
Expressions (4) and (7) will be evaluated for TW ≤ 1/λ.
4 Performance Comparison
In this section, we compare the power consumption
and delay performance of WiseMAC, PTIP and PSM as
a function of the protocol parameter TW . The choice
of TW permits to trade a higher delay against a lower
average power consumption.
The performance of the protocols is also influenced
by a number of system parameters. For the radio
transceiver parameters, we will consider those of the
WiseNETTM wireless sensor System-on-a-Chip (SoC)
developed at CSEM [20], as listed in Table 1. This SoC
Figure 4: Power consumption and delay of WiseMAC, PTIP
and PSM as a function of the wake-up period TW .
includes a low power FSK radio transceiver (dual-band
434 and 868 MHz) [21], the CoolRISC 8 bits low power
microcontroller core [22], a random access memory as
well as digital and analog interfaces. It is designed on
the 0.18 µm standard digital CMOS process, and op-
erates from 1.5 V to 0.9 V such that a single inexpen-
sive alkaline battery can be used as the energy source.
The remaining parameters have been chosen as follows:
The frequency tolerance of the quartz is chosen to be
θ = ±30 · 10−6 , which corresponds to an inaccuracy
of 2.6 seconds over one day. This value is conservative,
as 32.768 kHz crystals targeted for the wristwatch mar-
ket reach a frequency tolerance better than 20 · 10−6 at
25 ◦ C (see for example [23]). The length of the data and
of the control packets are chosen to be respectively 50
and 10 bytes. We consider N = 10 sensor nodes and an
inter-arrival per sensor node of L = 1000 s = 16.6 min.
Recall that this traffic is supposed to consist in configu-
ration and query commands sent by the sensor network
controller. Such large inter-arrivals make hence sense in
this context. The sensitivity of the results to variations
of the system parameters will be discussed in section 5.
Fig. 4 shows the power consumption P and the
delay D as a function of the protocol parameter TW .
The horizontal line in the upper plot represents the
power consumption of an ideal protocol, which is de-
fined as the power needed to power on, receive the
data, turn-around the transceiver and send the acknowl-
edgement. In this case, we have PIDEAL = PZ +
PbR (TS +TD +TT )+PbT TC
L = 5.12 µW. It can be noticed that
the incremental cost due to the data reception is only of
0.12 µW, a small value compared to PZ = 5 µW.
In the lower plot, the horizontal line represents the
 12
10
8 Assuming that, based on the requirements of some ap-
Power [µW]
6 WiseMAC, this would result in an average delay of 0.6
4 above the DOZE power consumption. When powered by
Ideal
2 WiseMAC
PTIP
PSM
0 −2
10 10
−1
10
0 1
10
2
10
3
10
Delay D [s]
Figure 5: Power-delay characteristics of WiseMAC, PTIP and
PSM.
minimum delay that would be obtained with an ideal
protocol, i.e. DIDEAL = TD = 16 ms. WiseMAC
and PSM can approach this limit for small values of TW ,
but at a high power cost. PTIP cannot provide such short
delays, as using a short TW would result in an excessive
polling traffic. For increasing values of TW , the three
curves converge to D = TW /2, i.e. the transmission de-
lay becomes negligible as compared to the waiting delay.
Using both plots, a designer can choose a protocol
and a value for the parameter TW , making a trade-off
between the consumed power and the average transmis-
sion delay. To compare the protocols between them, one
can combine both plots and draw the power as a func-
tion of the delay as shown in Fig. 5. The horizontal line
represents the ideal power consumption and the vertical
line the ideal delay. The power consumption of PSM is
drawn only up to a delay of 50 s. For larger values of the
delay, the assumption 1/λ ≥ TW would be violated.
With delays below 200 s, one can see that WiseMAC
consumes less power than PTIP and PSM. The cost of
receiving data being negligible when L = 1000 s, this
can be understood by comparing the basic cost of their
respective wake-up scheme. With an infinitely low traffic
(L → ∞), the power consumption of WiseMAC, PTIP
and PSM becomes
PbR (TS + 1/B)
lim PW iseM AC = PZ +
L→∞ TW
PbT TC + PbR (TS + TT + TC )
lim PP T IP = PZ +
L→∞ TW
PbR (TS + TC )
lim PP SM = PZ + 2θPbR +
L→∞ TW
With WiseMAC, the transceiver powers up every TW
to listen to the channel during the duration of a radio
symbol. With PTIP, the transceiver periodically sends
a POLL packet and receives a reply. With PSM, the
transceiver periodically receives a TIM packet. As the
duration of a TIM packet is always larger than the du-
ration of a modulation symbol, the wake-up scheme of
WiseMAC consumes less than the one of PSM. As re-
ceiving a TIM packet consumes less than transmitting a
POLL and receiving a reply, PSM consumes less than
PTIP.
plication, one would choose the value TW = 1 s. With
second and a power consumption of 7 µW, only 2 µW
a single alkaline battery of 2.6 Ah capacity with a con-
stant power leakage of 27 µW, this power consumption
would translate into a lifetime of 8 years without uplink
traffic. See [17] for a description of the battery model.
For the same delay of 0.6 second, the power consump-
tion of PSM amounts to 11 µW, i.e. 57% more than
WiseMAC. To consume 7 µW with PSM, the average
delay must be extended to 2 seconds using TW = 4 s.
With the PTIP protocol, a delay of more than 20 sec-
onds must be accepted to reach a low power consump-
tion of 7 µW. Although PTIP consumes more energy
than WiseMAC and PSM, it can be of interest because
it is extremely simple to implement. If some application
can tolerate a large transmission delay (e.g. automated
meter reading), the PTIP protocol can become an excel-
lent choice. It must also be noticed that uplink traffic
can be used to piggy-back POLL requests. If an appli-
cation requires periodic uplink transmissions, then the
PTIP protocol can be implemented for the downlink at
no cost.
5 Sensitivity Analysis
In the previous section, typical values have been cho-
sen for the system parameters. A sensitivity analysis has
been performed to understand the impact of the varia-
tions of those parameters. Details can be found in [18].
The main findings are listed below.
Impact of large inter-arrival and scalability: For
small values of L, the duration of the WiseMAC wake-up
preamble is smaller than the wake-up interval, we have
TP = min(4θL, TW ) = 4θL. Overhearing is mitigated
in a probabilistic way. For large values of L, we have
TP = min(4θL, TW ) = TW . Overhearing is not miti-
gated anymore, but with very large values of L, the im-
pact of overhearing becomes small. When TW = 4θL
(or L = T4θ W
= 8333 s = 2.3 hours), overhearing is
maximized. With only a few sensor nodes in a network,
this has no serious impact. WiseMAC remains extremely
energy efficient. However, as the overhearing compo-
nent is proportional to N − 1 (the number of potential
overhearers), WiseMAC presents a scalability limitation,
especially when the inter-arrival L is near T4θ
W
. A solu-
tion to reduce the overhearing when L is large consists
in repeating the data frame within the wake-up preamble.
Overhearers stop listening to the transmission as soon as
they have received a copy of the data frame, and realized
that it is destined to another node.
One can observe in expression (4) that the power con-
sumption of PSM is independent of N . The PSM proto-
col is perfectly scalable. This is made possible through
the combined use of the polling technique (which avoids
overhearing) and the regular broadcast of the traffic indi-
cation map (which avoids excessive useless polling traf-
fic). The PTIP protocol, on the other hand, is clearly
not scalable, the polling traffic being proportional to the
number of nodes.
Packet Size: It was seen that the incremental power
consumption caused by the transmission of data traffic is
negligible when L is large. The variation of the data
packet duration TD has hence little impact. Because
the power consumption of the PTIP and PSM wake-up
schemes depends on TC , an increase of the control pack-
ets duration would negatively impact PTIP and PSM.
Transceiver bit rate. With a high bit rate transceiver
such as a Lucent IEEE 802.11 PC-Card [24], the trans-
mission duration of a 10 bytes control packet is of
0.08 ms, while the setup time remains in the order of
1 ms. As a result, the power consumption of WiseMAC,
PTIP and PSM becomes nearly identical.
TX/RX Power Ratio. If the ratio between the power
consumption in TX and in RX states approaches 1, then
the power consumption of the PTIP protocol approaches
the power consumption of PSM.
6 Summary and Conclusion
This paper has proposed WiseMAC and PTIP for the
downlink of infrastructure wireless sensor networks. A
comparison was made with PSM, the power save pro-
tocol used in the IEEE 802.11 and IEEE 802.15.4 Zig-
Bee standards. The trade-off between power consump-
tion and delay was analyzed in low traffic conditions.
Analytical expressions were given to compute the power
and delay of each protocol, as a function of the wake-
up period. It was shown that WiseMAC provides, for
the same delay, a significantly lower power consumption
than PSM. When the wake-up period can be chosen to
be very large, it was seen that all protocols approach the
power consumption in DOZE state. In such a case, the
PTIP protocol becomes attractive as well, because of its
implementation simplicity.
REFERENCES
[1] G. Asada et al., “Wireless Integrated Network Sensors:
Low Power Systems on a Chip,” in Proc. European Solid
State Circuits Conf., 1998.
[2] J. Rabaey et al., “Picoradio supports ad hoc ultra-low
power wireless networking,” IEEE Computer Magazine,
pp. 42–48, July 2000.
[3] Chipcon, “CC1010 Integrated RF Transceiver and Micro-
controller,” Brochure, Sept 2002.
[4] A. Mainwaring, J. Polastre, R. Szewczyk, D. Culler,
and J. Anderson, “Wireless Sensor Networks for Habitat
Monitoring,” in Proc. Int. Workshop on Wireless Sensor
Networks and Applications (WSNA), Sept 2002.
[5] W. Ye, J. Heidemann, and D. Estrin, “An Energy-
Efficient MAC Protocol for Wireless Sensor Networks ,”
in Proc. IEEE INFOCOM Conf., 2002.
[6] L. Kleinrock and F. A. Tobagi, “Packet Switching in
Radio Channels: Part 1-Carrier Sense Multiple-Access
Modes and Their Throughput-Delay Characteristics,”
IEEE Transactions on Communications, vol. 23, no. 12,
pp. 1400–1416, 1975.
[7] A. El-Hoiydi, J.-D. Decotignie, C. Enz, and E. Le Roux,
“Poster Abstract: WiseMAC, An Ultra Low Power MAC
Protocol for the WiseNET Wireless Sensor Network,” in
Proc. 1st ACM SenSys Conf., Nov. 2003, pp. 302–303.
[8] A. Capone, M. Gerla, and R. R. Kapoor, “Efficient
polling schemes for Bluetooth picocells,” in Proc. IEEE
Int. Conf. on Communications, June 2001, pp. 1990–
1994.
[9] IEEE, “Wireless LAN Medium Access Control (MAC)
and Physical Layer (PHY) Specifications,” IEEE 802.11,
ISO/IEC 8802-11:1999, 1999.
[10] ——, “Wireless Medium Access Control (MAC) and
Physical Layer (PHY) specifications for Low Rate
Wireless Personal Area Networks (LR-WPANS),” IEEE
802.15.4-2003, 2003.
[11] V. Bharghavan et al., “MACAW: A Media Access Pro-
tocol for Wireless LAN’s,” in Proc. ACM SIGCOMM
Conf., 1994.
[12] J.-C. Chen, K. M. Sivalingam, and P. Agrawal, “Per-
formance Comparison of Battery Power Consumption in
Wireless Multiple Access Protocols,” Wireless Networks,
vol. 5, pp. 445–460, 1999.
[13] F. Bennett et al., “Piconet: embedded mobile network-
ing,” IEEE Journal on Personal Communications, vol. 4,
no. 5, pp. 8–15, Oct 1997.
[14] B. Mangione-Smith, “Low power communications pro-
tocols: paging and beyond ,” in Proc. IEEE Symposium
on Low Power Electronics, Oct. 1995, pp. 8–11.
[15] ITU-R, “M.584-2 Codes and formats for radio paging,”
Nov 1997.
[16] A. El-Hoiydi, “Aloha with Preamble Sampling for Spo-
radic Traffic in Ad Hoc Wireless Sensor Networks,” in
Proc. IEEE Int. Conf. on Communications, New York,
USA, Apr 2002, pp. 3418–3423.
[17] ——, “Spatial TDMA and CSMA with Preamble Sam-
pling for Low Power Ad Hoc Wireless Sensor Networks,”
in Proc. IEEE Int. Conf. on Computers and Communica-
tions (ISCC), Taormina, Italy, July 2002, pp. 685–692.
[18] A. El-Hoiydi, J.-D. Decotignie, and J. Hernandez, “Low
Power MAC Protocols for Infrastructure Wireless Sensor
Networks,” CSEM Technical Report, Nov. 27 2003.
[19] J. Capetanakis, “Tree algorithms for packet broadcast
channels,” IEEE Transactions on Information Theory,
vol. 25, no. 5, pp. 505 –515, Sep 1979.
[20] WiseNET, http://www.csem.ch/wisenet.
[21] F. Giroud, E. Le Roux, T. Melly, F. Pengg, V. Peiris,
N. Raemy, and D. Ruffieux, “WiseNET: Design of a
Low-Power RF CMOS Receiver Chip for Wireless Ap-
plications,” CSEM Scientific and Technical Report, p. 22,
2001.
[22] Xemics, “CoolRISC 816 8-bit Microprocessor Core -
Hardware and Software Reference Manual,” Version 4.5,
Apr 2001.
[23] SaRonix, “32.768 kHz Tubular Crystal NTF3238 /
NTF3226 Series,” Datasheet, DS-118 REV D.
[24] L. Technologies, “Orinoco Silver PC Card - Getting
Started Guide,” Aug 2000.